A Predictive Sample Reuse (PSR) approach is being developed for problems of Model Selection and Allocation (or diagnosis) for a wide variety of low and high structured statistical paradigms. This includes distribution-free procedures as well as binomial, exponential univariate and multivariate normal distribution assumptions. The case where both the model and the allocator are chosen simultaneously is also resolved in several different ways. The PSR approach is applied to growth curves in order to obtain optimal predictors and their assessment.